Conventional emotion-recognition approaches rely on group-averaged features, overlooking meaningful inter-individual variability in emotion regulation. We investigate within-emotion autonomic heterogeneity using heart rate variability (HRV) instantaneous Sympathetic and Parasympathetic Activity Indices (SAI, PAI) from a point-process electrocardiographic (ECG) framework in 38 participants undergoing a validated virtual reality protocol eliciting four primary emotions. Building on previously identified phenotypes, we quantify SAI/PAI temporal dynamics via cluster-based permutation. Fear showed the greatest heterogeneity: a fight-or-flight phenotype (24%) with progressive SAI increase and PAI collapse, versus a freeze phenotype (76%) with SAI suppression and preserved parasympathetic tone. Happiness, Sadness, and Relax each showed a majority (90–97%) and minority phenotype. A cluster-based permutation test revealed a sustained divergence in both SAI and PAI between the Fear phenotypes over most of the scene. These results show that autonomic phenotypes are distinguished not only by their average level but by a specific, temporally structured divergence, supporting phenotype-aware affective computing.